Model-based fault diagnosis techniques. Design schemes, algorithms and tools (Q694487)

The main subject of this book is the study of fault diagnosis. It includes methodologies to study fault detection, isolation, and identification. In the book the subject of threshold computation and residual generation is also addressed. Several methodologies used to study the problem of fault diagnosis are presented as well as many algorithms based on some of the techniques considered. This book consists of four parts. In the first part, we find a small introduction describing the aim of the book, some historic account of the theme considered, as well as some basic concepts and results related to fault detection and isolation. Some modeling methodologies are also presented. Among the many issues addressed in this first part of the book, are the description of nominal system behavior, coprime factorization technique, representation of systems with disturbances, modeling of faults, and some examples. The subjects of fault detectability, isolation, and identifiability are also taken into account in this first part. Among the many examples presented are speed control of a direct current motor, inverted pendulum control system, continuous stirred tank heater, and three-tank system. The second part of the book is dedicated to residual generation. This part contains four chapters. The first of them, is dedicated to the presentation of some basic residual generation methods. Among the subjects addressed in this first chapter are the concept of analytic redundancy, parametrization of residuals generators, fault detection filters, construction of diagnostic observer-based residual generators, and characterization of the so-called Luenberger equations. A numerical and an algebraic approaches to solving these equations are also given. Towards the end of the chapter, the authors presents the important model-based residual generation approach called parity space approach. Comparison and interconnections between different approaches are also discussed. The second chapter of this second part, deals with the generation of residuals that are decoupled from unknown inputs. The concept of perfect unknown input decoupling problem as well as condition for its existence are addressed. In order to verify the existence of such decoupling, the author considers the Rosenbrock system matrix as well as an algebraic method. Geometric and eigenstructure assignment approaches are considered to study the case of unknown input fault detection filter. In the case of unknown input diagnostic observer, the author presents an algebraic unknown input observer, the so-called matrix pencil, and numerical approaches to design such systems. Towards the end of this second chapter, further approaches to the design problem are presented and discussed. The third chapter of this second part, is dedicated to residual generation that are more robust against unknown inputs. After some preliminary results are given, the author starts presenting the methods considered, especially the Kalman filter, the so-called \({\mathcal H}_{2}/{\mathcal H}_{2}\) design, linear matrix inequality aided design, and the so-called \({\mathcal H}_{2}\) to \({\mathcal H}_{-}\) trade off design. In this chapter, the author also presents some indices of robustness, fault sensitivity and performance, as well as criteria for the optimal selection of parity matrices and vectors. At the end of the chapter, we find an index for evaluating, in practice, fault sensitivity and with that, a way of achieving an optimal design for the observer fault design. The last chapter of this second part, presents forms of residual generation that are robust against model uncertainties. One way of achieving such aim, is through transforming model uncertainties into unknown inputs. Another strategy explored in this chapter, is the so-called reference model based. The chapter also deals with residual generation for systems with polytopic uncertainties as well as for stochastically uncertain systems. In the third part of the book, the subject of residual evaluation and threshold computation is addressed. This part deals with a norm-based residual evaluation, a statistical based method, and an integration of both. In the norm-based evaluation, some standard evaluation functions are presented as well as some basic ideas of threshold setting and problem formulation. Computation of some specific threshold such as the threshold measuring the maximum change of the average energy level in response to the model uncertainties and unknown inputs, are also presented. Forms of generating the threshold of interest are also given in several cases. When it comes to the statistical-based method, evaluation criteria such as, the likelihood ratio, hypothesis tests, and the detection of changes in the variance, are presented. Regarding the threshold computation, the Neyman-Pearson criterion as well as the maximum \textit{à posteriori} and the Bayes criteria, are considered. The author also presents the extended results to the solution of fault detection problem present in linear dynamics systems. In the last chapter of this third part, the author shows how to integrate the norm- and the statistical-based methods in order to study the fault detection and isolation problem. Hence, ways of evaluating residuals in a stochastic system with deterministic disturbances as well as in stochastically uncertain systems, are presented. The chapter also deals with threshold computation via probabilistic techniques. The last part of the book is dedicated to fault detection, isolation, and identification schemes. This part is split into four chapters. The first chapter is dedicated to the design of model-based fault detection systems. It deals with residual generation, definition of residual evaluation function, and threshold computation. The problem of maximizing fault detection when we have a given false alarm rate is addressed as well as the minimization of the false alarm rate under a given fault detection rate. Applications to stochastic systems are given towards the end of the chapter. The second chapter of this fourth part, is dedicated to studying the fault isolation problem. After some basic introduction is given, the author presents several approaches to the design of fault isolation filters. Those approaches are somehow a little restrictive. Hence, the author presents some less restrictive approaches such as, an algebraic one. A fault isolation scheme using the so-called bank of residual generation is also presented. After dealing with fault isolation, the author moves on to the fault identification problem. This identification problem is considered the main aim in fault diagnosis studies. One of the methodologies addressed in the book is the fault identification filter schemes. The author also considers the problem of optimality of fault identification filter design, where several approaches to the problem are given. Finally, in the last chapter, the subject of fault diagnosis in feedback control systems is addressed as well as the subject of fault tolerant architectures. Among the type of models considered are, the so-called plat and control loop models and the parametrization of stabilizing controllers and observers. We also have a study of how to extend the residual in the standard feedback control loop and also a presentation of a fault detection scheme. The author also develops the theme of accessing residuals in the internal model control. The case of control structures based on residual generator is also considered. The book contains several examples illustrating the use of many of the methodologies presented as well as the pseudo-code of algorithms of the several procedures discussed. However, in a future edition, in the same manner as we benefit from a list of some of the notation used, the reader would benefit from a list of acronyms used throughout the work. That list could be placed either at the beginning or at the end of the book.